contextd serve

Start contextd as an MCP (Model Context Protocol) server.

Usage

contextd serve [options]

Options

Flag	Description	Default
`--port <port>`	Port to serve on	`3333`

What it does

Running contextd serve starts an MCP server over stdio transport that exposes your project context as tools AI assistants can call:

Tool	Description	Parameters
`get_project_overview`	Get high-level project overview + architecture	None
`get_conventions`	Get coding conventions and standards	None
`get_relevant_context`	Get context relevant to specific files	`files` — array of file paths
`list_decisions`	List all architecture decision records	None
`get_module_context`	Get context for a specific module	`module` — module name or path

Startup behavior

Finds the project root
Auto-refreshes stale remote sources (>24 hours old) non-blocking
Warns on very stale sources (>7 days)
Dynamically imports @modelcontextprotocol/sdk
Creates MCP server with name contextd and version 0.1.0
Connects via stdio transport
Prints MCP configuration JSON to stderr

MCP configuration

The server prints this configuration to stderr when started:

{
  "mcpServers": {
    "contextd": {
      "command": "npx",
      "args": ["contextd", "serve"]
    }
  }
}

Add it to .mcp.json in your project root, or to your MCP client’s settings.

Remote context inclusion

All tools automatically include remote contexts if configured. When get_relevant_context is called, the server matches files against both local and cached remote module scopes.

Decision Auth